pcre_exec.c

来自「Ubuntu packages of security software。 相」· C语言 代码 · 共 2,090 行 · 第 1/5 页

#define rdepth             frame->Xrdepth

/* Ditto for the local variables */

#ifdef SUPPORT_UTF8
#define charptr            frame->Xcharptr
#endif
#define callpat            frame->Xcallpat
#define data               frame->Xdata
#define next               frame->Xnext
#define pp                 frame->Xpp
#define prev               frame->Xprev
#define saved_eptr         frame->Xsaved_eptr

#define new_recursive      frame->Xnew_recursive

#define cur_is_word        frame->Xcur_is_word
#define condition          frame->Xcondition
#define minimize           frame->Xminimize
#define prev_is_word       frame->Xprev_is_word

#define original_ims       frame->Xoriginal_ims

#ifdef SUPPORT_UCP
#define prop_type          frame->Xprop_type
#define prop_value         frame->Xprop_value
#define prop_fail_result   frame->Xprop_fail_result
#define prop_category      frame->Xprop_category
#define prop_chartype      frame->Xprop_chartype
#define prop_script        frame->Xprop_script
#define prop_test_variable frame->Xprop_test_variable
#endif

#define ctype              frame->Xctype
#define fc                 frame->Xfc
#define fi                 frame->Xfi
#define length             frame->Xlength
#define max                frame->Xmax
#define min                frame->Xmin
#define number             frame->Xnumber
#define offset             frame->Xoffset
#define op                 frame->Xop
#define save_capture_last  frame->Xsave_capture_last
#define save_offset1       frame->Xsave_offset1
#define save_offset2       frame->Xsave_offset2
#define save_offset3       frame->Xsave_offset3
#define stacksave          frame->Xstacksave

#define newptrb            frame->Xnewptrb

/* When recursion is being used, local variables are allocated on the stack and
get preserved during recursion in the normal way. In this environment, fi and
i, and fc and c, can be the same variables. */

#else
#define fi i
#define fc c


#ifdef SUPPORT_UTF8                /* Many of these variables are used only  */
const uschar *charptr;             /* in small blocks of the code. My normal */
#endif                             /* style of coding would have declared  */
const uschar *callpat;             /* them within each of those blocks.    */
const uschar *data;                /* However, in order to accommodate the */
const uschar *next;                /* version of this code that uses an    */
USPTR         pp;                  /* external "stack" implemented on the    */
const uschar *prev;                /* heap, it is easier to declare them all */
USPTR         saved_eptr;          /* here, so the declarations can be cut   */
                                   /* out in a block. The only declarations  */
recursion_info new_recursive;      /* within blocks below are for variables  */
                                   /* that do not have to be preserved over  */
BOOL cur_is_word;                  /* a recursive call to RMATCH().          */
BOOL condition;
BOOL minimize;
BOOL prev_is_word;

unsigned long int original_ims;

#ifdef SUPPORT_UCP
int prop_type;
int prop_value;
int prop_fail_result;
int prop_category;
int prop_chartype;
int prop_script;
int *prop_test_variable;
#endif

int ctype;
int length;
int max;
int min;
int number;
int offset;
int op;
int save_capture_last;
int save_offset1, save_offset2, save_offset3;
int stacksave[REC_STACK_SAVE_MAX];

eptrblock newptrb;
#endif

/* These statements are here to stop the compiler complaining about unitialized
variables. */

#ifdef SUPPORT_UCP
prop_value = 0;
prop_fail_result = 0;
prop_test_variable = NULL;
#endif

/* This label is used for tail recursion, which is used in a few cases even
when NO_RECURSE is not defined, in order to reduce the amount of stack that is
used. Thanks to Ian Taylor for noticing this possibility and sending the
original patch. */

TAIL_RECURSE:

/* OK, now we can get on with the real code of the function. Recursive calls
are specified by the macro RMATCH and RRETURN is used to return. When
NO_RECURSE is *not* defined, these just turn into a recursive call to match()
and a "return", respectively (possibly with some debugging if DEBUG is
defined). However, RMATCH isn't like a function call because it's quite a
complicated macro. It has to be used in one particular way. This shouldn't,
however, impact performance when true recursion is being used. */

/* First check that we haven't called match() too many times, or that we
haven't exceeded the recursive call limit. */

if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);

original_ims = ims;    /* Save for resetting on ')' */

#ifdef SUPPORT_UTF8
utf8 = md->utf8;       /* Local copy of the flag */
#else
utf8 = FALSE;
#endif

/* At the start of a bracketed group, add the current subject pointer to the
stack of such pointers, to be re-instated at the end of the group when we hit
the closing ket. When match() is called in other circumstances, we don't add to
this stack. */

if ((flags & match_isgroup) != 0)
  {
  newptrb.epb_prev = eptrb;
  newptrb.epb_saved_eptr = eptr;
  eptrb = &newptrb;
  }

/* Now start processing the operations. */

for (;;)
  {
  op = *ecode;
  minimize = FALSE;

  /* For partial matching, remember if we ever hit the end of the subject after
  matching at least one subject character. */

  if (md->partial &&
      eptr >= md->end_subject &&
      eptr > md->start_match)
    md->hitend = TRUE;

  /* Opening capturing bracket. If there is space in the offset vector, save
  the current subject position in the working slot at the top of the vector. We
  mustn't change the current values of the data slot, because they may be set
  from a previous iteration of this group, and be referred to by a reference
  inside the group.

  If the bracket fails to match, we need to restore this value and also the
  values of the final offsets, in case they were set by a previous iteration of
  the same bracket.

  If there isn't enough space in the offset vector, treat this as if it were a
  non-capturing bracket. Don't worry about setting the flag for the error case
  here; that is handled in the code for KET. */

  if (op > OP_BRA)
    {
    number = op - OP_BRA;

    /* For extended extraction brackets (large number), we have to fish out the
    number from a dummy opcode at the start. */

    if (number > EXTRACT_BASIC_MAX)
      number = GET2(ecode, 2+LINK_SIZE);
    offset = number << 1;

#ifdef DEBUG
    printf("start bracket %d subject=", number);
    pchars(eptr, 16, TRUE, md);
    printf("\n");
#endif

    if (offset < md->offset_max)
      {
      save_offset1 = md->offset_vector[offset];
      save_offset2 = md->offset_vector[offset+1];
      save_offset3 = md->offset_vector[md->offset_end - number];
      save_capture_last = md->capture_last;

      DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
      md->offset_vector[md->offset_end - number] = eptr - md->start_subject;

      do
        {
        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
          match_isgroup);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        md->capture_last = save_capture_last;
        ecode += GET(ecode, 1);
        }
      while (*ecode == OP_ALT);

      DPRINTF(("bracket %d failed\n", number));

      md->offset_vector[offset] = save_offset1;
      md->offset_vector[offset+1] = save_offset2;
      md->offset_vector[md->offset_end - number] = save_offset3;

      RRETURN(MATCH_NOMATCH);
      }

    /* Insufficient room for saving captured contents */

    else op = OP_BRA;
    }

  /* Other types of node can be handled by a switch */

  switch(op)
    {
    case OP_BRA:     /* Non-capturing bracket: optimized */
    DPRINTF(("start bracket 0\n"));

    /* Loop for all the alternatives */

    for (;;)
      {
      /* When we get to the final alternative within the brackets, we would
      return the result of a recursive call to match() whatever happened. We
      can reduce stack usage by turning this into a tail recursion. */

      if (ecode[GET(ecode, 1)] != OP_ALT)
      {
       ecode += 1 + LINK_SIZE;
       flags = match_isgroup;
       DPRINTF(("bracket 0 tail recursion\n"));
       goto TAIL_RECURSE;
       }

      /* For non-final alternatives, continue the loop for a NOMATCH result;
      otherwise return. */

      RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
        match_isgroup);
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
      ecode += GET(ecode, 1);
      }
    /* Control never reaches here. */

    /* Conditional group: compilation checked that there are no more than
    two branches. If the condition is false, skipping the first branch takes us
    past the end if there is only one branch, but that's OK because that is
    exactly what going to the ket would do. As there is only one branch to be
    obeyed, we can use tail recursion to avoid using another stack frame. */

    case OP_COND:
    if (ecode[LINK_SIZE+1] == OP_CREF) /* Condition extract or recurse test */
      {
      offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
      condition = (offset == CREF_RECURSE * 2)?
        (md->recursive != NULL) :
        (offset < offset_top && md->offset_vector[offset] >= 0);
      ecode += condition? (LINK_SIZE + 4) : (LINK_SIZE + 1 + GET(ecode, 1));
      flags = match_isgroup;
      goto TAIL_RECURSE;
      }

    /* The condition is an assertion. Call match() to evaluate it - setting
    the final argument TRUE causes it to stop at the end of an assertion. */

    else
      {
      RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
          match_condassert | match_isgroup);
      if (rrc == MATCH_MATCH)
        {
        ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE+2);
        while (*ecode == OP_ALT) ecode += GET(ecode, 1);
        }
      else if (rrc != MATCH_NOMATCH)
        {
        RRETURN(rrc);         /* Need braces because of following else */
        }
      else ecode += GET(ecode, 1);

      /* We are now at the branch that is to be obeyed. As there is only one,
      we can use tail recursion to avoid using another stack frame. */

      ecode += 1 + LINK_SIZE;
      flags = match_isgroup;
      goto TAIL_RECURSE;
      }
    /* Control never reaches here */

    /* Skip over conditional reference or large extraction number data if
    encountered. */

    case OP_CREF:
    case OP_BRANUMBER:
    ecode += 3;
    break;

    /* End of the pattern. If we are in a recursion, we should restore the
    offsets appropriately and continue from after the call. */

    case OP_END:
    if (md->recursive != NULL && md->recursive->group_num == 0)
      {
      recursion_info *rec = md->recursive;
      DPRINTF(("End of pattern in a (?0) recursion\n"));
      md->recursive = rec->prevrec;
      memmove(md->offset_vector, rec->offset_save,
        rec->saved_max * sizeof(int));
      md->start_match = rec->save_start;
      ims = original_ims;
      ecode = rec->after_call;
      break;
      }

    /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty
    string - backtracking will then try other alternatives, if any. */

    if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);
    md->end_match_ptr = eptr;          /* Record where we ended */
    md->end_offset_top = offset_top;   /* and how many extracts were taken */
    RRETURN(MATCH_MATCH);

    /* Change option settings */

    case OP_OPT:
    ims = ecode[1];
    ecode += 2;
    DPRINTF(("ims set to %02lx\n", ims));
    break;

    /* Assertion brackets. Check the alternative branches in turn - the
    matching won't pass the KET for an assertion. If any one branch matches,
    the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
    start of each branch to move the current point backwards, so the code at
    this level is identical to the lookahead case. */

    case OP_ASSERT:
    case OP_ASSERTBACK:
    do
      {
      RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
        match_isgroup);
      if (rrc == MATCH_MATCH) break;
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
      ecode += GET(ecode, 1);
      }
    while (*ecode == OP_ALT);
    if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);

    /* If checking an assertion for a condition, return MATCH_MATCH. */

    if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);

    /* Continue from after the assertion, updating the offsets high water
    mark, since extracts may have been taken during the assertion. */

    do ecode += GET(ecode,1); while (*ecode == OP_ALT);
    ecode += 1 + LINK_SIZE;
    offset_top = md->end_offset_top;
    continue;

    /* Negative assertion: all branches must fail to match */

    case OP_ASSERT_NOT:
    case OP_ASSERTBACK_NOT:
    do
      {
      RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
        match_isgroup);
      if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
      ecode += GET(ecode,1);
      }
    while (*ecode == OP_ALT);

    if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);

    ecode += 1 + LINK_SIZE;
    continue;

    /* Move the subject pointer back. This occurs only at the start of
    each branch of a lookbehind assertion. If we are too close to the start to
    move back, this match function fails. When working with UTF-8 we move
    back a number of characters, not bytes. */

    case OP_REVERSE:
#ifdef SUPPORT_UTF8
    if (utf8)
      {
      c = GET(ecode,1);
      for (i = 0; i < c; i++)
        {
        eptr--;
        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
        BACKCHAR(eptr)
        }
      }